Structure Self-Adaptive 3d Model Editing Method

1. A structure self-adaptive three-dimensional (3D) model editing method, comprising: step (S 100 ), clustering 3D models of a same category according to structures: inputting 3D models of the same category, clustering the 3D models into different groups according to difference in components contained in the 3D models, wherein structures of models in the same group are required to be as similar as possible and structures of models between different groups are required to be as different as possible; step (S 200 ), learning a design knowledge prior of intra-group 3D models: compiling statistics of relationships between the components of models in the same group using a multivariate linear regression model according to a result of the clustering of the 3D models of the same category, to guide a 3D model editing procedure with structures thereof being preserved; step (S 300 ), learning a structure switching rule of inter-group 3D models: analyzing geometrical parameter distribution of common components of the models in different groups according to the result of the clustering of the 3D models of the same category, and obtaining the structure switching rule of inter-group 3D models; and step (S 400 ), optimizing a user-edited 3D model: editing, by a user using an interactive tool, a size, a position, and/or an angle parameter of a component of a 3D model, adjusting a structure of the user-edited 3D model and automatically optimizing geometrical parameters of other components of the user-edited 3D model according to the learned design knowledge prior of intra-group 3D models and the learned structure switching rule of inter-group 3D models so that the optimized 3D model satisfies a design knowledge prior of a model library, the structure self-adaptive 3D model editing method is used to increase a 3D model editing speed and improve degree of automation of the model editing.

2. The method according to claim 1 , wherein the clustering 3D models of a same category according to structures in step (S 100 ) comprises: normalizing sizes and positions of 3D models in the model library, and pre-dividing the models in the model library into a component level, wherein the method —does not require a semantic corresponding relationship between components of different models, the corresponding relationship between components of different models is obtained automatically by clustering the components according to a position relationship between the components; after clustering the components of the 3D models, determining a quantity N of component types in the models of the same category and defining a set of the components as {P 1 , P 2 , . . . , P N }, for the model S i in the model library, a vector x i containing N elements is obtained, if the model S i contains a component P n , then x i (n)=1, otherwise, x i (n)=0, given any two 3D models S i and S j , vectors x i and x j are obtained, wherein a distance ( ) between the two vectors is defined as: ( x i , x j ) = ∑ ( n 1 · n 2 ) ⁢ x i ⁡ ( n 1 ) ⁢ ( 1 ⁢ ⁢ x j ⁡ ( n 1 ) ) ⁢ ( 1 ⁢ ⁢ x i ⁡ ( n 2 ) ) ⁢ x j ⁡ ( n 21 ) ( 1 max ⁡ ( φ ⁡ ( n 1 , n 2 ) , φ ⁡ ( n 1 , n 2 ) ) ) wherein x i (n 1 ) (1x j (n 1 ))(1x i (n 2 ))x j (n 21 ) is used to determine the quantity of component types in the two 3D models, φ(n 1 , n 2 ) is conditional probability, and is presented as φ(n 1 ,n 2 )=P(x(n)=1) that is co-existent probability of two components; and after obtaining the distance between vectors of two random 3D models, obtaining a distance matrix by calculating the distance between each two vectors of 3D models in a random model library, and realizing the clustering of 3D models via a spectral clustering algorithm.

3. The method according to claim 1 , wherein the learning a design knowledge prior of intra-group 3D models in step (S 200 ) comprises: obtaining an orientated bounding box (OBB) for a component of a 3D model, which comprises following three steps: first, obtaining an approximate convex hull for coordinate points of the 3D model, taking a random plane on the approximate convex hull as a projection plane and projecting all the points of the model to the projection plane, calculating a 2D OBB of the projected points and stretching the 2D OBB along a plane normal direction until all the points of the model are included, which forms a candidate OBB; then, calculating a quantity of symmetric planes of respective OBBs, defining three candidate planes for any OBB i, wherein these candidate planes are determined by a center C i and three axial directions (a i 1 , a i 2 , a i 3 ) of the OBB, obtaining uniform sampling points on a surface of the model, calculating reflected points of the sampling points when reflected by a random candidate plane, calculating distances from these reflected points to the surface of the model, if a distance is smaller than 0.0001, determining the corresponding sampling point as a symmetric point, if a percentage of the symmetric points exceeds 90%, determining the candidate plane as a symmetric plane; eventually, determining an optimal OBB which has the most symmetric planes, if a plurality of candidate OBBs contain a same quantity of symmetric planes, determining the OBB with a smallest volume as a final OBB; extracting parameters of the component of the 3D model, given a random model, OBBs of a plurality of components are obtained, for component j of model i, a center C j i of the OBB, three axial directions (a j,1 i , a j,2 i , a j,3 i ), and the lengths (e j,1 i , e j,2 i , e j,3 i ) of the OBB in the respective axial directions are obtained, these parameters are used to extract nine-dimensional parameters, for the component of model i, parameters F j i =(f j,1 i ,f j,2 i ,f j,3 i . . . f j,9 i ) are obtained, wherein first three parameters represent the center of the OBB, middle three parameters represent projection angles between the respective axial directions and corresponding world coordinate axes, and last three parameters represent lengths of the OBB on the three directions, the obtained nine-dimensional parameters are successively used in the learning of the design knowledge prior of the model, which are inputs of an intra-group design knowledge prior learning module, and are also candidate threshold values of an inter-group structure switching parameter of an inter-group design knowledge prior learning module; and learning an intra-group design knowledge prior, the intra-group design knowledge prior emphasizes on learning a deformation rule of the model while the structure thereof is preserved, in a same structure group, defining a component that all the modules contain as a common component, a total quantity of the common components is M, defining a multivariate regression coefficient matrix as {A (m) |m=1, . . . M}, wherein A (m) contains regression coefficients of all the common components, a i =α 0 , . . . α n represents an i-th row of the matrix, which is computed as: a i = arg ⁢ ⁢ min a i ⁢  b i ⁢ ⁢ α 0 ⁢ ⁢ ∑ j ≠ 1 ⁢ ⁢ α j ⁢ ⁢ b j  2 2 wherein b 1 , . . . b n are parameters of all the common components, by repeating the above computing process, a final regression coefficient matrix A m is obtained; the formula can only solve a relationship between the common components, for a private component of the model, a relationship is constructed by establishing a linear regression equation between parameters of a private component and the common component, for a parameter b of the private component, a parameter {circumflex over (b)} of the common component which has a highest correlation coefficient with the parameter b is selected, and then unary linear regression equations {circumflex over (b)}=β 1 b+β 0 and b={circumflex over (β)} 1 {circumflex over (b)}+{circumflex over (β)} 0 are calculated as limit deformation equations.

4. The method according to claim 1 , wherein the learning a structure switching rule of inter-group 3D models in step (S 300 ) comprises: determining a candidate threshold value of structure switching, given two structure groups S i and S j , wherein the two structure groups contain components having a corresponding relationship, then a parameter of a common component can be threshold values of structure switching; and determining a final threshold value of structure switching, using an M×M matrix to represent a structure switching rule of inter-group 3D models, the matrix is expressed as {T n t |t=1, . . . 9} n=1 N , wherein T n t represents a correlative value of any two structure groups with respective to a t-th parameter of a component P n , T n t (i,j) represents whether the parameter should be the threshold value between the structure groups S i and S j , assuming the component P n is the common component of the structure groups S i and S j , the t-th parameters of the components P n of all the models in the two structure groups is represented as {B i |b 1 i , . . . b N i } and {B j |b 1 j , . . . b N j }, then d(B) and d(b, B) are defined as: d ⁡ ( B ) = max i , j ⁢ (  b i ⁢ ⁢ b j  ) d ⁡ ( b , B ) = min b i , ∈ B ⁢ (  b ⁢ ⁢ b i  ) then a definition of T n t (i, j) is obtained as: T n t ⁡ ( i , j ) = { λ , if ⁢ ⁢ max b n j ∈ B j ⁢ ⁢ ( b n j , B i ) > λ , ∀ i ≠ j ⁢ ∞ , ot ⁢ ⁢ erwise wherein λ = d ⁡ ( B i ) + d ⁡ ( B j ) 2 , if max b n j εB j (b n j ,B i )>λ, then the parameter is used as a threshold value.

5. The method according to claim 1 , wherein the optimizing a user-edited 3D model in step (S 400 ) comprises: editing a component of a 3D model, wherein a user selects a component of a 3D model using a mouse, and conducts translation, rotation, scaling, deleting and adding operations to the corresponding component; performing self-adaptive structure conversion: if the user selects the component of the model and operates a parameter, then setting the parameter as b, if the parameter is a threshold value of inter-group structure switching and a current structure group is S i , the following formula is used to determine if the structure needs to be converted to S j : K ⁡ ( b ) = { 1 , if ( b , B j ) ≤ T n t ⁡ ( i , j ) ⁢ ⁢ and ⁢ ⁢ ( b , B i ) < T n t ⁡ ( i , j ) 0 , ot erwise if K(b)=1, the structure needs to be converted from S i to S j , otherwise the structure is preserved; and optimizing a component parameter of a 3D model with a structure thereof being preserved, inputting parameters of all common components into a trained multivariate regression model, and obtaining estimated values of respective component parameters to maintain rationality of model design, assuming a current structure group is S m , a trained multivariate regression coefficient matrix is A m , and assuming the component parameters of the current 3D model are b=[b 1 , . . . b n ], if the user converts a parameter b c to B 0 , then all the remaining component parameters are calculated as: arg b ⁢ min ⁢ ∑ i = 1 n ⁢ ⁢  b i ⁢ ⁢ ( ω i ⁢ b i ′ + ( 1 ⁢ ⁢ ω i ) ⁢ b ~ i )  2 2 + ω c ⁢  b c ⁢ ⁢ B 0  2 2 wherein {tilde over (b)} i is an initial value of b i , b i ′ is a multivariate regression estimated value of b i , and ω i is a multivariate regression determination coefficient, wherein ω c ∥b c B 0 ∥ 2 2 prevents the user-edited component parameter from being affected by a result of the multivariate regression model, where ω c =10.